基于环境特征有效提取的机械手末端位姿嵌入式控制研究

Research on Embedded Control of Robot Arm End Pose Based on Effective Extraction of Environmental Features

摘要:

在柔性机械手的作业环境中,光照变化与物体表面反射特性之间具有非线性,待抓取物体表面复杂的形状和纹理特征也呈现出非线性,严重影响了对环境特征的提取效果,导致最终的控制精度偏低。对此,本研究利用双RBF神经网络具有的局部逼近特性和泛化学习能力,设计针对作业环境特征的提取方法。同时,以嵌入式的方式设计位姿控制器,针对末端位姿嵌入式控制方法展开研究。首先,利用内置传感器检测柔性机械手末端的初始位姿；然后,针对采集到的机械手作业环境图像实施预处理,再利用双RBF神经网络提取复杂的、非线性的环境特征。通过特征匹配确定作业目标,并将其作为控制目标。最后,以当前位姿为起点、以控制目标为终点,规划机械手末端移动轨迹,根据实际位姿与轨迹点之间的偏差确定位姿控制量,实现位姿控制。根据测试可知:与传统方法相比,应用本文方法控制后,机械手末端位置控制误差明显降低,姿态角控制误差减小。

Abstract:

In the working environment of flexible robotic arms, there is a non-linear relationship between changes in lighting and the reflection characteristics of object surfaces. The complex shape and texture features of the object surface to be grasped also exhibit non-linearity, which seriously affects the extraction effect of environmental features and leads to low final control accuracy. In response to this, this study utilizes the local approximation properties and generalization learning ability of the dual RBF neural network to design a feature extraction method for the work environment. At the same time, design a pose controller in an embedded manner and conduct research on embedded control methods for end effector poses. Firstly, use built-in sensors to detect the initial pose of the end effector of the flexible robotic arm; Then, preprocessing is applied to the collected images of the robotic arm operating environment, and a dual RBF neural network is used to extract complex and nonlinear environmental features. Determine the task objective through feature matching and use it as the control objective. Finally, starting from the current pose and ending with the control target, plan the end effector movement trajectory of the robotic arm, and determine the pose control quantity based on the deviation between the actual pose and the trajectory points to achieve pose control. According to the test, it can be seen that compared with traditional methods, the application of the method proposed in this article significantly reduces the end position control error of the robotic arm and the attitude angle control error.